
import librosa
import librosa.display
import numpy as np
import matplotlib.pyplot as plt


from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive


import librosa
audio_data = '/content/drive/MyDrive/Classroom/Projects/Music Genre Classification/Data/genres_original/blues/blues.00006.wav'
x , sr = librosa.load(audio_data)
print(type(x), type(sr))

<class 'numpy.ndarray'> <class 'int'>


librosa.load(audio_data, sr=44100)

(array([ 0.18108444,  0.26436776,  0.28416514, ..., -0.47086203,
        -0.4199038 , -0.21476781], dtype=float32), 44100)


import IPython.display as ipd
ipd.Audio(audio_data)


%matplotlib inline
import matplotlib.pyplot as plt
import librosa.display
plt.figure(figsize=(14, 5))
librosa.display.waveplot(x, sr=sr)

<matplotlib.collections.PolyCollection at 0x7fd6ff473910>


def to_decibles(signal):
    # Perform short time Fourier Transformation of signal and take absolute value of results
    stft = np.abs(librosa.stft(signal))
    # Convert to dB
    D = librosa.amplitude_to_db(stft, ref = np.max) # Set reference value to the maximum value of stft.
    return D # Return converted audio signal

# Function to plot the converted audio signal
def plot_spec(D, sr, instrument):
    fig, ax = plt.subplots(figsize = (30,10))
    spec = librosa.display.specshow(D, sr=sr, x_axis='time', y_axis='linear', ax=ax)
    ax.set(title = 'Spectrogram of ' + instrument)
    fig.colorbar(spec)


plot_spec(to_decibles(x), sr, 'Blues')


from PIL import Image
im = Image.open('/content/drive/MyDrive/Classroom/Projects/Music Genre Classification/Data/images_original/blues/blues00006.png')
im.show()


import IPython.display as display
from PIL import Image
display.display(Image.open('/content/drive/MyDrive/Classroom/Projects/Music Genre Classification/Data/images_original/blues/blues00006.png'))


from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive


# Ignore  the warnings
import warnings
warnings.filterwarnings('always')
warnings.filterwarnings('ignore')

# data visualisation and manipulation
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib import style
import seaborn as sns
 
#configure
# sets matplotlib to inline and displays graphs below the corressponding cell.
%matplotlib inline  
style.use('fivethirtyeight')
sns.set(style='whitegrid',color_codes=True)

#model selection
from sklearn.model_selection import train_test_split
from sklearn.model_selection import KFold
from sklearn.metrics import accuracy_score,precision_score,recall_score,confusion_matrix,roc_curve,roc_auc_score
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import LabelEncoder

#preprocess.
from keras.preprocessing.image import ImageDataGenerator

#dl libraraies
from keras import backend as K
from keras.models import Sequential
from keras.layers import Dense
from tensorflow.keras.optimizers import Adam, SGD, RMSprop,Adagrad,Adadelta
from tensorflow.keras.utils import to_categorical
from keras.callbacks import ReduceLROnPlateau

# specifically for cnn
from keras.layers import Dropout, Flatten,Activation
from keras.layers import Conv2D, MaxPooling2D, BatchNormalization
 
import tensorflow as tf
import random as rn

# specifically for manipulating zipped images and getting numpy arrays of pixel values of images.
import cv2                  
import numpy as np  
from tqdm import tqdm
import os                   
from random import shuffle  
from zipfile import ZipFile
from PIL import Image

#TL pecific modules
from keras.applications.vgg16 import VGG16


import os
print(os.listdir('/content/drive/MyDrive/Classroom/Projects/Music Genre Classification/Data/images_original'))

['metal', 'blues', 'classical', 'pop', 'jazz', 'country', 'reggae', 'rock', 'disco', 'hiphop']


data = pd.read_csv('/content/drive/MyDrive/Classroom/Projects/Music Genre Classification/Data/features_30_sec.csv')


def get_image_path(x):
    splited = x.split('.')
    folder = '/content/drive/MyDrive/Classroom/Projects/Music Genre Classification/Data/images_original/'+splited[0] + '/'
    filename = splited[0]+splited[1]+'.png'
    return folder+filename


data['path'] = data['filename'].apply(get_image_path)
data


data = data[['filename','path']]


data['genre'] = data['filename'].apply(lambda x:x.split('.')[0])
data


X = data.iloc[: ,0:1]
y = data.iloc[:,1]


from sklearn.model_selection import train_test_split
X_train,y_train, X_test,y_test = train_test_split(X,y,  test_size=0.2, random_state=15)


len(X_train.genre.unique()),len(X_test.genre.unique())

(10, 10)


X_train.genre.hist()

<matplotlib.axes._subplots.AxesSubplot at 0x7fdb148a3a10>


X_test.genre.hist()

<matplotlib.axes._subplots.AxesSubplot at 0x7fdb147c3ed0>


from tensorflow.keras.preprocessing.image import ImageDataGenerator
datagen = ImageDataGenerator()


it = datagen.flow_from_dataframe(
    data,
    x_col="path",
    y_col="genre",batch_size=1)

Found 999 validated image filenames belonging to 10 classes.


import matplotlib.pyplot as plt
plt.imshow(next(it)[0][0])

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

<matplotlib.image.AxesImage at 0x7fdbc170bed0>


next(it)[0][0].shape

(256, 256, 3)


from tensorflow.keras.applications import ResNet50
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten,Dense,Input
inp = Input(shape = next(it)[0][0].shape)
model = ResNet50(
    include_top=False,
    weights="imagenet",
    
    classes=10,
    input_tensor = inp
)

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
94773248/94765736 [==============================] - 1s 0us/step
94781440/94765736 [==============================] - 1s 0us/step


for layer in model.layers:
    layer.trainable=False


model.input.shape

TensorShape([None, 256, 256, 3])


from tensorflow.keras.layers import BatchNormalization,Dropout,Conv2D
new_model = Sequential()
new_model.add(model)
new_model.add(Flatten())
new_model.add(BatchNormalization())
new_model.add(Dense(128, activation='relu'))
new_model.add(Dropout(0.5))
new_model.add(BatchNormalization())
new_model.add(Dense(64, activation='relu'))
new_model.add(Dropout(0.5))
new_model.add(Dense(10, activation='softmax'))


new_model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 resnet50 (Functional)       (None, 8, 8, 2048)        23587712  
                                                                 
 flatten (Flatten)           (None, 131072)            0         
                                                                 
 batch_normalization (BatchN  (None, 131072)           524288    
 ormalization)                                                   
                                                                 
 dense (Dense)               (None, 128)               16777344  
                                                                 
 dropout (Dropout)           (None, 128)               0         
                                                                 
 batch_normalization_1 (Batc  (None, 128)              512       
 hNormalization)                                                 
                                                                 
 dense_1 (Dense)             (None, 64)                8256      
                                                                 
 dropout_1 (Dropout)         (None, 64)                0         
                                                                 
 dense_2 (Dense)             (None, 10)                650       
                                                                 
=================================================================
Total params: 40,898,762
Trainable params: 17,048,650
Non-trainable params: 23,850,112
_________________________________________________________________


new_model.compile(loss = 'categorical_crossentropy',optimizer='Adam',metrics=['accuracy'])


from tensorflow.keras.callbacks import EarlyStopping
es = EarlyStopping(
    monitor='val_accuracy', restore_best_weights=True, patience=3
)

new_model.fit(datagen.flow_from_dataframe(
    X_train,
    x_col="path",
    y_col="genre",batch_size=100),
              validation_data=datagen.flow_from_dataframe(
            X_test,
            x_col="path",
            y_col="genre",batch_size=64),epochs=50,callbacks=[es])

Found 799 validated image filenames belonging to 10 classes.
Found 200 validated image filenames belonging to 10 classes.
Epoch 1/50
8/8 [==============================] - 224s 26s/step - loss: 2.4331 - accuracy: 0.2253 - val_loss: 3.9933 - val_accuracy: 0.4000
Epoch 2/50
8/8 [==============================] - 9s 1s/step - loss: 1.8723 - accuracy: 0.3692 - val_loss: 2.4654 - val_accuracy: 0.4700
Epoch 3/50
8/8 [==============================] - 9s 1s/step - loss: 1.6537 - accuracy: 0.4518 - val_loss: 1.9852 - val_accuracy: 0.4800
Epoch 4/50
8/8 [==============================] - 9s 1s/step - loss: 1.4796 - accuracy: 0.4981 - val_loss: 1.7590 - val_accuracy: 0.5000
Epoch 5/50
8/8 [==============================] - 9s 1s/step - loss: 1.3779 - accuracy: 0.5369 - val_loss: 1.6236 - val_accuracy: 0.5250
Epoch 6/50
8/8 [==============================] - 9s 1s/step - loss: 1.2338 - accuracy: 0.5857 - val_loss: 1.5007 - val_accuracy: 0.5350
Epoch 7/50
8/8 [==============================] - 9s 1s/step - loss: 1.1374 - accuracy: 0.6170 - val_loss: 1.4129 - val_accuracy: 0.5400
Epoch 8/50
8/8 [==============================] - 9s 1s/step - loss: 1.0221 - accuracy: 0.6621 - val_loss: 1.3530 - val_accuracy: 0.5400
Epoch 9/50
8/8 [==============================] - 9s 1s/step - loss: 0.9196 - accuracy: 0.7222 - val_loss: 1.3047 - val_accuracy: 0.5600
Epoch 10/50
8/8 [==============================] - 9s 1s/step - loss: 0.8680 - accuracy: 0.7171 - val_loss: 1.2710 - val_accuracy: 0.5650
Epoch 11/50
8/8 [==============================] - 9s 1s/step - loss: 0.7620 - accuracy: 0.7572 - val_loss: 1.2480 - val_accuracy: 0.5800
Epoch 12/50
8/8 [==============================] - 9s 1s/step - loss: 0.6739 - accuracy: 0.8035 - val_loss: 1.2330 - val_accuracy: 0.5850
Epoch 13/50
8/8 [==============================] - 9s 1s/step - loss: 0.6426 - accuracy: 0.8098 - val_loss: 1.2164 - val_accuracy: 0.5900
Epoch 14/50
8/8 [==============================] - 9s 1s/step - loss: 0.5666 - accuracy: 0.8486 - val_loss: 1.1884 - val_accuracy: 0.6200
Epoch 15/50
8/8 [==============================] - 9s 1s/step - loss: 0.5161 - accuracy: 0.8511 - val_loss: 1.1655 - val_accuracy: 0.6200
Epoch 16/50
8/8 [==============================] - 9s 1s/step - loss: 0.4488 - accuracy: 0.8773 - val_loss: 1.1443 - val_accuracy: 0.6500
Epoch 17/50
8/8 [==============================] - 9s 1s/step - loss: 0.4216 - accuracy: 0.8899 - val_loss: 1.1349 - val_accuracy: 0.6300
Epoch 18/50
8/8 [==============================] - 9s 1s/step - loss: 0.3358 - accuracy: 0.9199 - val_loss: 1.1392 - val_accuracy: 0.6300
Epoch 19/50
8/8 [==============================] - 9s 1s/step - loss: 0.3272 - accuracy: 0.9287 - val_loss: 1.1423 - val_accuracy: 0.6250

<keras.callbacks.History at 0x7fdb100c66d0>


new_model.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 2s 372ms/step - loss: 1.1443 - accuracy: 0.6500

[1.1443243026733398, 0.6499999761581421]


ResNet_scores = new_model.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 2s 370ms/step - loss: 1.1443 - accuracy: 0.6500


print(" %.2f%%" % ( ResNet_scores[1]*100))

 65.00%


it = datagen.flow_from_dataframe(
    data,
    x_col="path",
    y_col="genre",batch_size=1)

Found 999 validated image filenames belonging to 10 classes.


from tensorflow.keras.applications import VGG16
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten,Dense,Input
inp = Input(shape = next(it)[0][0].shape)
model1 = VGG16(
    include_top=False,
    weights="imagenet",
    
    classes=10,
    input_tensor = inp
)

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58892288/58889256 [==============================] - 0s 0us/step
58900480/58889256 [==============================] - 0s 0us/step


for layer in model1.layers:
    layer.trainable=False


model1.input.shape

TensorShape([None, 256, 256, 3])


from tensorflow.keras import layers
# Flatten the output layer to 1 dimension
new_model1 = layers.Flatten()(model1.output)

# Add a fully connected layer with 512 hidden units and ReLU activation
new_model1 = layers.Dense(4096, activation='relu')(new_model1)

# Add a dropout rate of 0.5
new_model1 = layers.Dropout(0.5)(new_model1)

# Add a final sigmoid layer with 1 node for classification output
new_model1 = layers.Dense(10, activation='softmax')(new_model1)

final_model1 = tf.keras.models.Model(model1.input, new_model1)


final_model1.summary()

Model: "model"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_2 (InputLayer)        [(None, 256, 256, 3)]     0         
                                                                 
 block1_conv1 (Conv2D)       (None, 256, 256, 64)      1792      
                                                                 
 block1_conv2 (Conv2D)       (None, 256, 256, 64)      36928     
                                                                 
 block1_pool (MaxPooling2D)  (None, 128, 128, 64)      0         
                                                                 
 block2_conv1 (Conv2D)       (None, 128, 128, 128)     73856     
                                                                 
 block2_conv2 (Conv2D)       (None, 128, 128, 128)     147584    
                                                                 
 block2_pool (MaxPooling2D)  (None, 64, 64, 128)       0         
                                                                 
 block3_conv1 (Conv2D)       (None, 64, 64, 256)       295168    
                                                                 
 block3_conv2 (Conv2D)       (None, 64, 64, 256)       590080    
                                                                 
 block3_conv3 (Conv2D)       (None, 64, 64, 256)       590080    
                                                                 
 block3_pool (MaxPooling2D)  (None, 32, 32, 256)       0         
                                                                 
 block4_conv1 (Conv2D)       (None, 32, 32, 512)       1180160   
                                                                 
 block4_conv2 (Conv2D)       (None, 32, 32, 512)       2359808   
                                                                 
 block4_conv3 (Conv2D)       (None, 32, 32, 512)       2359808   
                                                                 
 block4_pool (MaxPooling2D)  (None, 16, 16, 512)       0         
                                                                 
 block5_conv1 (Conv2D)       (None, 16, 16, 512)       2359808   
                                                                 
 block5_conv2 (Conv2D)       (None, 16, 16, 512)       2359808   
                                                                 
 block5_conv3 (Conv2D)       (None, 16, 16, 512)       2359808   
                                                                 
 block5_pool (MaxPooling2D)  (None, 8, 8, 512)         0         
                                                                 
 flatten (Flatten)           (None, 32768)             0         
                                                                 
 dense (Dense)               (None, 4096)              134221824 
                                                                 
 dropout (Dropout)           (None, 4096)              0         
                                                                 
 dense_1 (Dense)             (None, 10)                40970     
                                                                 
=================================================================
Total params: 148,977,482
Trainable params: 134,262,794
Non-trainable params: 14,714,688
_________________________________________________________________


final_model1.compile(loss = 'categorical_crossentropy',optimizer='Adam',metrics=['accuracy'])


X_train.shape

(800, 3)


from tensorflow.keras.callbacks import EarlyStopping
es = EarlyStopping(
    monitor='val_accuracy', restore_best_weights=True, patience=3
)

final_model1.fit(datagen.flow_from_dataframe(
    X_train,
    x_col="path",
    y_col="genre",batch_size=100),
              validation_data=datagen.flow_from_dataframe(
            X_test,
            x_col="path",
            y_col="genre",batch_size=64),epochs=50,callbacks=[es])

Found 799 validated image filenames belonging to 10 classes.
Found 200 validated image filenames belonging to 10 classes.
Epoch 1/50
8/8 [==============================] - 36s 5s/step - loss: 272.9385 - accuracy: 0.2215 - val_loss: 171.4277 - val_accuracy: 0.2500
Epoch 2/50
8/8 [==============================] - 15s 2s/step - loss: 107.1405 - accuracy: 0.4230 - val_loss: 45.4078 - val_accuracy: 0.4150
Epoch 3/50
8/8 [==============================] - 16s 2s/step - loss: 24.8963 - accuracy: 0.5419 - val_loss: 24.4607 - val_accuracy: 0.4750
Epoch 4/50
8/8 [==============================] - 15s 2s/step - loss: 10.1852 - accuracy: 0.6771 - val_loss: 13.1776 - val_accuracy: 0.5200
Epoch 5/50
8/8 [==============================] - 15s 2s/step - loss: 4.1752 - accuracy: 0.7572 - val_loss: 7.4210 - val_accuracy: 0.6100
Epoch 6/50
8/8 [==============================] - 15s 2s/step - loss: 1.9307 - accuracy: 0.8123 - val_loss: 5.3333 - val_accuracy: 0.5800
Epoch 7/50
8/8 [==============================] - 15s 2s/step - loss: 0.8089 - accuracy: 0.8798 - val_loss: 4.7627 - val_accuracy: 0.6250
Epoch 8/50
8/8 [==============================] - 15s 2s/step - loss: 0.5196 - accuracy: 0.9111 - val_loss: 3.7698 - val_accuracy: 0.6200
Epoch 9/50
8/8 [==============================] - 15s 2s/step - loss: 0.5053 - accuracy: 0.9186 - val_loss: 3.5156 - val_accuracy: 0.5900
Epoch 10/50
8/8 [==============================] - 15s 2s/step - loss: 0.1546 - accuracy: 0.9574 - val_loss: 2.8220 - val_accuracy: 0.6450
Epoch 11/50
8/8 [==============================] - 15s 2s/step - loss: 0.1317 - accuracy: 0.9599 - val_loss: 2.9224 - val_accuracy: 0.6250
Epoch 12/50
8/8 [==============================] - 15s 2s/step - loss: 0.0711 - accuracy: 0.9775 - val_loss: 3.1377 - val_accuracy: 0.6150
Epoch 13/50
8/8 [==============================] - 15s 2s/step - loss: 0.0834 - accuracy: 0.9787 - val_loss: 2.8900 - val_accuracy: 0.6300

<keras.callbacks.History at 0x7fda17efa8d0>


final_model1.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 3s 638ms/step - loss: 2.8220 - accuracy: 0.6450

[2.8220410346984863, 0.6449999809265137]


VGG16_scores = final_model1.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 3s 518ms/step - loss: 2.8220 - accuracy: 0.6450


print(VGG16_scores)

[2.822039842605591, 0.6449999809265137]


print(" %.2f%%" % ( VGG16_scores[1]*100))

 64.50%


it = datagen.flow_from_dataframe(data,x_col="path",y_col="genre",batch_size=1)

Found 999 validated image filenames belonging to 10 classes.


from tensorflow.keras.applications import InceptionV3
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten,Dense,Input
inp = Input(shape = next(it)[0][0].shape)
model2 = InceptionV3(
    include_top=False,
    weights="imagenet",
    classes=10,
    input_tensor = inp
)

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/inception_v3/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5
87916544/87910968 [==============================] - 1s 0us/step
87924736/87910968 [==============================] - 1s 0us/step


for layer in model2.layers:
    layer.trainable=False


from tensorflow.keras import layers
# Flatten the output layer to 1 dimension
new_model2 = layers.Flatten()(model2.output)

# Add a fully connected layer with 512 hidden units and ReLU activation
new_model2 = layers.Dense(1024, activation='relu')(new_model2)

# Add a dropout rate of 0.2
new_model2 = layers.Dropout(0.2)(new_model2)

# Add a final sigmoid layer with 1 node for classification output
new_model2 = layers.Dense(10, activation='softmax')(new_model2)

final_model2 = tf.keras.models.Model(model2.input, new_model2)


final_model2.summary()

Model: "sequential_6"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 inception_v3 (Functional)   (None, 6, 6, 2048)        21802784  
                                                                 
 flatten_5 (Flatten)         (None, 73728)             0         
                                                                 
 dense_7 (Dense)             (None, 1024)              75498496  
                                                                 
 dropout_3 (Dropout)         (None, 1024)              0         
                                                                 
 dense_8 (Dense)             (None, 10)                10250     
                                                                 
=================================================================
Total params: 97,311,530
Trainable params: 75,508,746
Non-trainable params: 21,802,784
_________________________________________________________________


final_model2.compile(loss = 'categorical_crossentropy',optimizer='Adam',metrics=['accuracy'])


from tensorflow.keras.callbacks import EarlyStopping
es = EarlyStopping(
    monitor='val_accuracy', restore_best_weights=True, patience=3
)

final_model2.fit(datagen.flow_from_dataframe(
    X_train,
    x_col="path",
    y_col="genre",batch_size=100),
              validation_data=datagen.flow_from_dataframe(
            X_test,
            x_col="path",
            y_col="genre",batch_size=64),epochs=50,callbacks=[es])

Found 799 validated image filenames belonging to 10 classes.
Found 200 validated image filenames belonging to 10 classes.
Epoch 1/50
8/8 [==============================] - 24s 2s/step - loss: 2312.7266 - accuracy: 0.1189 - val_loss: 1369.4250 - val_accuracy: 0.2100
Epoch 2/50
8/8 [==============================] - 8s 967ms/step - loss: 1123.6622 - accuracy: 0.2153 - val_loss: 705.7430 - val_accuracy: 0.2600
Epoch 3/50
8/8 [==============================] - 8s 962ms/step - loss: 383.7527 - accuracy: 0.2904 - val_loss: 246.4879 - val_accuracy: 0.2800
Epoch 4/50
8/8 [==============================] - 8s 984ms/step - loss: 161.1486 - accuracy: 0.3304 - val_loss: 74.0731 - val_accuracy: 0.3350
Epoch 5/50
8/8 [==============================] - 8s 959ms/step - loss: 35.6517 - accuracy: 0.4080 - val_loss: 9.7351 - val_accuracy: 0.3550
Epoch 6/50
8/8 [==============================] - 8s 916ms/step - loss: 4.2483 - accuracy: 0.3217 - val_loss: 28.9307 - val_accuracy: 0.1050
Epoch 7/50
8/8 [==============================] - 8s 924ms/step - loss: 5.2831 - accuracy: 0.2603 - val_loss: 2.0551 - val_accuracy: 0.2700
Epoch 8/50
8/8 [==============================] - 8s 959ms/step - loss: 2.6576 - accuracy: 0.2728 - val_loss: 2.0020 - val_accuracy: 0.2900

<keras.callbacks.History at 0x7fda9935f150>


final_model2.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 2s 327ms/step - loss: 9.7351 - accuracy: 0.3550

[9.735074043273926, 0.35499998927116394]


InceptionV3_scores = final_model2.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 2s 299ms/step - loss: 9.7351 - accuracy: 0.3550


print(" %.2f%%" % ( InceptionV3_scores[1]*100))

 35.50%


it = datagen.flow_from_dataframe(data,x_col="path",y_col="genre",batch_size=1)

Found 999 validated image filenames belonging to 10 classes.


from tensorflow.keras.applications import Xception
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten,Dense,Input
inp = Input(shape = next(it)[0][0].shape)
model3 = Xception(
    include_top=False,
    weights="imagenet",
    
    classes=10,
    input_tensor = inp
)

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/xception/xception_weights_tf_dim_ordering_tf_kernels_notop.h5
83689472/83683744 [==============================] - 1s 0us/step
83697664/83683744 [==============================] - 1s 0us/step


for layer in model3.layers:
    layer.trainable=False


from tensorflow.keras import layers
# Flatten the output layer to 1 dimension
new_model3 = layers.Flatten()(model3.output)

# Add a dropout rate of 0.2
new_model3 = layers.Dropout(0.2)(new_model3)

# Add a final sigmoid layer with 1 node for classification output
new_model3 = layers.Dense(10, activation='softmax')(new_model3)

final_model3 = tf.keras.models.Model(model3.input, new_model3)


final_model3.summary()

Model: "sequential_7"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 xception (Functional)       (None, 8, 8, 2048)        20861480  
                                                                 
 flatten_6 (Flatten)         (None, 131072)            0         
                                                                 
 dropout_4 (Dropout)         (None, 131072)            0         
                                                                 
 dense_9 (Dense)             (None, 10)                1310730   
                                                                 
=================================================================
Total params: 22,172,210
Trainable params: 1,310,730
Non-trainable params: 20,861,480
_________________________________________________________________


final_model3.compile(loss = 'categorical_crossentropy',optimizer='Adam',metrics=['accuracy'])


from tensorflow.keras.callbacks import EarlyStopping
es = EarlyStopping(
    monitor='val_accuracy', restore_best_weights=True, patience=3
)

final_model3.fit(datagen.flow_from_dataframe(
    X_train,
    x_col="path",
    y_col="genre",batch_size=100),
              validation_data=datagen.flow_from_dataframe(
            X_test,
            x_col="path",
            y_col="genre",batch_size=64),epochs=50,callbacks=[es])

Found 799 validated image filenames belonging to 10 classes.
Found 200 validated image filenames belonging to 10 classes.
Epoch 1/50
8/8 [==============================] - 72s 9s/step - loss: 183.0119 - accuracy: 0.1339 - val_loss: 242.0247 - val_accuracy: 0.0850
Epoch 2/50
8/8 [==============================] - 12s 2s/step - loss: 144.4788 - accuracy: 0.1677 - val_loss: 103.8414 - val_accuracy: 0.1900
Epoch 3/50
8/8 [==============================] - 12s 1s/step - loss: 78.9643 - accuracy: 0.2904 - val_loss: 67.3496 - val_accuracy: 0.2600
Epoch 4/50
8/8 [==============================] - 12s 1s/step - loss: 40.9229 - accuracy: 0.3442 - val_loss: 41.3923 - val_accuracy: 0.3000
Epoch 5/50
8/8 [==============================] - 12s 1s/step - loss: 23.7086 - accuracy: 0.4318 - val_loss: 37.0942 - val_accuracy: 0.2650
Epoch 6/50
8/8 [==============================] - 12s 1s/step - loss: 17.7745 - accuracy: 0.4831 - val_loss: 28.7289 - val_accuracy: 0.3150
Epoch 7/50
8/8 [==============================] - 12s 1s/step - loss: 15.8346 - accuracy: 0.4856 - val_loss: 23.3874 - val_accuracy: 0.3850
Epoch 8/50
8/8 [==============================] - 12s 1s/step - loss: 10.4950 - accuracy: 0.5469 - val_loss: 23.2457 - val_accuracy: 0.4200
Epoch 9/50
8/8 [==============================] - 12s 2s/step - loss: 8.1085 - accuracy: 0.6045 - val_loss: 17.3948 - val_accuracy: 0.4400
Epoch 10/50
8/8 [==============================] - 13s 2s/step - loss: 5.9844 - accuracy: 0.6283 - val_loss: 17.5081 - val_accuracy: 0.4550
Epoch 11/50
8/8 [==============================] - 14s 2s/step - loss: 5.1647 - accuracy: 0.6621 - val_loss: 17.1475 - val_accuracy: 0.4500
Epoch 12/50
8/8 [==============================] - 12s 1s/step - loss: 5.9869 - accuracy: 0.6558 - val_loss: 16.6585 - val_accuracy: 0.4300
Epoch 13/50
8/8 [==============================] - 13s 2s/step - loss: 5.0638 - accuracy: 0.6984 - val_loss: 17.6007 - val_accuracy: 0.4250

<keras.callbacks.History at 0x7fda1eb10950>


final_model3.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 3s 500ms/step - loss: 17.5081 - accuracy: 0.4550

[17.508089065551758, 0.45500001311302185]


Xception_scores = final_model3.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 3s 500ms/step - loss: 17.5081 - accuracy: 0.4550


print(" %.2f%%" % ( Xception_scores[1]*100))

 45.50%


it = datagen.flow_from_dataframe(data,x_col="path",
    y_col="genre",batch_size=1)

Found 999 validated image filenames belonging to 10 classes.


from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten,Dense,Input
inp = Input(shape = next(it)[0][0].shape)
model4 = MobileNetV2(
    include_top=False,
    weights="imagenet",
    
    classes=10,
    input_tensor = inp
)


for layer in model4.layers:
    layer.trainable=False


from tensorflow.keras.layers import BatchNormalization,Dropout
new_model4 = Sequential()
new_model4.add(model4)
new_model4.add(Flatten())
new_model4.add(Dense(1024, activation='relu'))
new_model4.add(Dropout(0.5))
new_model4.add(BatchNormalization())
new_model4.add(Dense(10, activation='softmax'))


new_model4.summary()

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 mobilenetv2_1.00_224 (Funct  (None, 8, 8, 1280)       2257984   
 ional)                                                          
                                                                 
 flatten_16 (Flatten)        (None, 81920)             0         
                                                                 
 dense_30 (Dense)            (None, 1024)              83887104  
                                                                 
 dropout_16 (Dropout)        (None, 1024)              0         
                                                                 
 batch_normalization_288 (Ba  (None, 1024)             4096      
 tchNormalization)                                               
                                                                 
 dense_31 (Dense)            (None, 10)                10250     
                                                                 
=================================================================
Total params: 86,159,434
Trainable params: 83,899,402
Non-trainable params: 2,260,032
_________________________________________________________________


new_model4.compile(loss = 'categorical_crossentropy',optimizer='Adam',metrics=['accuracy'])


from tensorflow.keras.callbacks import EarlyStopping
es = EarlyStopping(
    monitor='val_accuracy', restore_best_weights=True, patience=3
)

new_model4.fit(datagen.flow_from_dataframe(
    X_train,
    x_col="path",
    y_col="genre",batch_size=100),
              validation_data=datagen.flow_from_dataframe(
            X_test,
            x_col="path",
            y_col="genre",batch_size=64),epochs=50,callbacks=[es])

Found 799 validated image filenames belonging to 10 classes.
Found 200 validated image filenames belonging to 10 classes.
Epoch 1/50
8/8 [==============================] - 18s 1s/step - loss: 2.1871 - accuracy: 0.3517 - val_loss: 7.3380 - val_accuracy: 0.3850
Epoch 2/50
8/8 [==============================] - 9s 1s/step - loss: 1.3493 - accuracy: 0.5620 - val_loss: 3.6017 - val_accuracy: 0.4250
Epoch 3/50
8/8 [==============================] - 10s 1s/step - loss: 1.1329 - accuracy: 0.6033 - val_loss: 2.7677 - val_accuracy: 0.4400
Epoch 4/50
8/8 [==============================] - 11s 1s/step - loss: 0.9868 - accuracy: 0.6395 - val_loss: 2.1923 - val_accuracy: 0.4600
Epoch 5/50
8/8 [==============================] - 7s 899ms/step - loss: 0.8779 - accuracy: 0.6946 - val_loss: 1.8398 - val_accuracy: 0.4950
Epoch 6/50
8/8 [==============================] - 8s 1s/step - loss: 0.7249 - accuracy: 0.7672 - val_loss: 1.7551 - val_accuracy: 0.5000
Epoch 7/50
8/8 [==============================] - 8s 1s/step - loss: 0.5648 - accuracy: 0.8210 - val_loss: 1.6029 - val_accuracy: 0.5150
Epoch 8/50
8/8 [==============================] - 12s 1s/step - loss: 0.4745 - accuracy: 0.8548 - val_loss: 1.4889 - val_accuracy: 0.5450
Epoch 9/50
8/8 [==============================] - 11s 1s/step - loss: 0.4120 - accuracy: 0.8861 - val_loss: 1.4957 - val_accuracy: 0.5350
Epoch 10/50
8/8 [==============================] - 7s 873ms/step - loss: 0.2910 - accuracy: 0.9387 - val_loss: 1.4238 - val_accuracy: 0.5300
Epoch 11/50
8/8 [==============================] - 7s 894ms/step - loss: 0.2485 - accuracy: 0.9362 - val_loss: 1.4743 - val_accuracy: 0.5300

<keras.callbacks.History at 0x7fda16e20510>


MobilenetV2_scores = new_model4.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 1s 298ms/step - loss: 1.4889 - accuracy: 0.5450


print("%.2f%%" %(MobilenetV2_scores[1]*100))

54.50%


it = datagen.flow_from_dataframe(
    data,
    x_col="path",
    y_col="genre",batch_size=1)

Found 999 validated image filenames belonging to 10 classes.


from tensorflow.keras.applications import DenseNet201
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten,Dense,Input
inp = Input(shape = next(it)[0][0].shape)
model5 = DenseNet201(
    include_top=False,
    weights="imagenet",
    
    classes=10,
    input_tensor = inp
)


for layer in model5.layers:
    layer.trainable=False


from tensorflow.keras.layers import BatchNormalization,Dropout,Conv2D
new_model5 = Sequential()
new_model5.add(model5)
new_model5.add(Flatten())
new_model5.add(BatchNormalization())
new_model5.add(Dense(1024, activation='relu'))
new_model5.add(Dropout(0.5))
new_model5.add(BatchNormalization())
new_model5.add(Dense(1024, activation='relu'))
new_model5.add(Dropout(0.5))
new_model5.add(Dense(10, activation='softmax'))


new_model5.summary()

Model: "sequential_6"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 densenet201 (Functional)    (None, 8, 8, 1920)        18321984  
                                                                 
 flatten_21 (Flatten)        (None, 122880)            0         
                                                                 
 batch_normalization_294 (Ba  (None, 122880)           491520    
 tchNormalization)                                               
                                                                 
 dense_36 (Dense)            (None, 1024)              125830144 
                                                                 
 dropout_19 (Dropout)        (None, 1024)              0         
                                                                 
 batch_normalization_295 (Ba  (None, 1024)             4096      
 tchNormalization)                                               
                                                                 
 dense_37 (Dense)            (None, 1024)              1049600   
                                                                 
 dropout_20 (Dropout)        (None, 1024)              0         
                                                                 
 dense_38 (Dense)            (None, 10)                10250     
                                                                 
=================================================================
Total params: 145,707,594
Trainable params: 127,137,802
Non-trainable params: 18,569,792
_________________________________________________________________


new_model5.compile(loss = 'categorical_crossentropy',optimizer='Adam',metrics=['accuracy'])


from tensorflow.keras.callbacks import EarlyStopping
es = EarlyStopping(
    monitor='val_accuracy', restore_best_weights=True, patience=3
)

new_model5.fit(datagen.flow_from_dataframe(
    X_train,
    x_col="path",
    y_col="genre",batch_size=100),
              validation_data=datagen.flow_from_dataframe(
            X_test,
            x_col="path",
            y_col="genre",batch_size=64),epochs=50,callbacks=[es])

Found 799 validated image filenames belonging to 10 classes.
Found 200 validated image filenames belonging to 10 classes.
Epoch 1/50
8/8 [==============================] - 60s 5s/step - loss: 2.1220 - accuracy: 0.3479 - val_loss: 32.4449 - val_accuracy: 0.1750
Epoch 2/50
8/8 [==============================] - 14s 2s/step - loss: 1.4588 - accuracy: 0.5156 - val_loss: 43.8013 - val_accuracy: 0.1100
Epoch 3/50
8/8 [==============================] - 16s 2s/step - loss: 1.2434 - accuracy: 0.6020 - val_loss: 20.3484 - val_accuracy: 0.2050
Epoch 4/50
8/8 [==============================] - 14s 2s/step - loss: 1.2181 - accuracy: 0.6158 - val_loss: 17.8662 - val_accuracy: 0.1600
Epoch 5/50
8/8 [==============================] - 13s 2s/step - loss: 0.9776 - accuracy: 0.6796 - val_loss: 14.7218 - val_accuracy: 0.1750
Epoch 6/50
8/8 [==============================] - 14s 2s/step - loss: 0.8473 - accuracy: 0.6934 - val_loss: 11.2352 - val_accuracy: 0.1750

<keras.callbacks.History at 0x7fda1b6de2d0>


DenseNet201_scores = new_model5.evaluate(datagen.flow_from_dataframe(
    X_test,
    x_col="path",
    y_col="genre",batch_size=64))

Found 200 validated image filenames belonging to 10 classes.
4/4 [==============================] - 3s 560ms/step - loss: 20.3484 - accuracy: 0.2050


print("%.2f%%" %(DenseNet201_scores[1]*100))

20.50%


from tabulate import tabulate
col_names = ['Model','Accuracy','Depth']
data = [['ResNet50',ResNet_scores[1]*100,107],
        ['VGG16',VGG16_scores[1]*100,16],
        ['InceptionV3',InceptionV3_scores[1]*100,189],
        ['Xception',Xception_scores[1]*100,81],
        ['MobileNetV2',MobilenetV2_scores[1]*100,105],
        ['DensetNet201',DenseNet201_scores[1]*100,402]]

print(tabulate(data, headers=col_names, tablefmt="fancy_grid", showindex="always", floatfmt=".2f"))

╒════╤══════════════╤════════════╤═════════╕
│    │ Model        │   Accuracy │   Depth │
╞════╪══════════════╪════════════╪═════════╡
│  0 │ ResNet50     │      65.00 │     107 │
├────┼──────────────┼────────────┼─────────┤
│  1 │ VGG16        │      64.50 │      16 │
├────┼──────────────┼────────────┼─────────┤
│  2 │ InceptionV3  │      35.50 │     189 │
├────┼──────────────┼────────────┼─────────┤
│  3 │ Xception     │      45.50 │      81 │
├────┼──────────────┼────────────┼─────────┤
│  4 │ MobileNetV2  │      54.50 │     105 │
├────┼──────────────┼────────────┼─────────┤
│  5 │ DensetNet201 │      20.50 │     402 │
╘════╧══════════════╧════════════╧═════════╛


import matplotlib.pyplot as plt
models=['ResNet50','VGG16','InceptionV3','Xception','MobileNet','DenseNet']
accuracy=[ResNet_scores[1]*100,VGG16_scores[1]*100,InceptionV3_scores[1]*100,
          Xception_scores[1]*100,MobilenetV2_scores[1]*100,DenseNet201_scores[1]*100]
fig, ax = plt.subplots()
plt.plot(models,accuracy,'b-o',label='Models Accuracy');
plt.xlabel('Models')
plt.ylabel('Accuracy')
plt.show()


import matplotlib.pyplot as plt
x = [ResNet_scores[1]*100,VGG16_scores[1]*100,InceptionV3_scores[1]*100,
          Xception_scores[1]*100,MobilenetV2_scores[1]*100,DenseNet201_scores[1]*100]
labels = ['ResNet50','VGG16','InceptionV3','Xception','MobileNetV2','DenseNet201']
fig, ax = plt.subplots()
ax.pie(x, labels = labels)
ax.set_title('Model Accuracies')
plt.show()


models=['ResNet50','VGG16','Inception','Xception','MobileNet','DenseNet']
depth = [ 107,16,189,81,105,402]
# Change color of each bar
plt.bar(models,depth, color=['firebrick', 'green', 'black', 'red',
                    'purple', 'skyblue'])
plt.xlabel('Models')
plt.ylabel("Depth")
plt.title('Depth of the Models')
plt.show()

	filename	length	chroma_stft_mean	chroma_stft_var	rms_mean	rms_var	spectral_centroid_mean	spectral_centroid_var	spectral_bandwidth_mean	spectral_bandwidth_var	...	mfcc17_mean	mfcc17_var	mfcc18_mean	mfcc18_var	mfcc19_mean	mfcc19_var	mfcc20_mean	mfcc20_var	label	path
0	blues.00000.wav	661794	0.350088	0.088757	0.130228	0.002827	1784.165850	129774.064525	2002.449060	85882.761315	...	-1.690215	36.524071	-0.408979	41.597103	-2.303523	55.062923	1.221291	46.936035	blues	/content/drive/MyDrive/Classroom/Projects/Musi...
1	blues.00001.wav	661794	0.340914	0.094980	0.095948	0.002373	1530.176679	375850.073649	2039.036516	213843.755497	...	-0.731125	60.314529	0.295073	48.120598	-0.283518	51.106190	0.531217	45.786282	blues	/content/drive/MyDrive/Classroom/Projects/Musi...
2	blues.00002.wav	661794	0.363637	0.085275	0.175570	0.002746	1552.811865	156467.643368	1747.702312	76254.192257	...	-7.729093	47.639427	-1.816407	52.382141	-3.439720	46.639660	-2.231258	30.573025	blues	/content/drive/MyDrive/Classroom/Projects/Musi...
3	blues.00003.wav	661794	0.404785	0.093999	0.141093	0.006346	1070.106615	184355.942417	1596.412872	166441.494769	...	-3.319597	50.206673	0.636965	37.319130	-0.619121	37.259739	-3.407448	31.949339	blues	/content/drive/MyDrive/Classroom/Projects/Musi...
4	blues.00004.wav	661794	0.308526	0.087841	0.091529	0.002303	1835.004266	343399.939274	1748.172116	88445.209036	...	-5.454034	75.269707	-0.916874	53.613918	-4.404827	62.910812	-11.703234	55.195160	blues	/content/drive/MyDrive/Classroom/Projects/Musi...
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
995	rock.00095.wav	661794	0.352063	0.080487	0.079486	0.000345	2008.149458	282174.689224	2106.541053	88609.749506	...	-13.289984	41.754955	2.484145	36.778877	-6.713265	54.866825	-1.193787	49.950665	rock	/content/drive/MyDrive/Classroom/Projects/Musi...
996	rock.00096.wav	661794	0.398687	0.075086	0.076458	0.000588	2006.843354	182114.709510	2068.942009	82426.016726	...	-10.848309	39.395096	1.881229	32.010040	-7.461491	39.196327	-2.795338	31.773624	rock	/content/drive/MyDrive/Classroom/Projects/Musi...
997	rock.00097.wav	661794	0.432142	0.075268	0.081651	0.000322	2077.526598	231657.968040	1927.293153	74717.124394	...	-12.845291	36.367264	3.440978	36.001110	-12.588070	42.502201	-2.106337	29.865515	rock	/content/drive/MyDrive/Classroom/Projects/Musi...
998	rock.00098.wav	661794	0.362485	0.091506	0.083860	0.001211	1398.699344	240318.731073	1818.450280	109090.207161	...	-4.416050	43.583942	1.556207	34.331261	-5.041897	47.227180	-3.590644	41.299088	rock	/content/drive/MyDrive/Classroom/Projects/Musi...
999	rock.00099.wav	661794	0.358401	0.085884	0.054454	0.000336	1609.795082	422203.216152	1797.213044	120115.632927	...	-7.069775	73.760391	0.028346	76.504326	-2.025783	72.189316	1.155239	49.662510	rock	/content/drive/MyDrive/Classroom/Projects/Musi...

	filename	path	genre
0	blues.00000.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	blues
1	blues.00001.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	blues
2	blues.00002.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	blues
3	blues.00003.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	blues
4	blues.00004.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	blues
...	...	...	...
995	rock.00095.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	rock
996	rock.00096.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	rock
997	rock.00097.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	rock
998	rock.00098.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	rock
999	rock.00099.wav	/content/drive/MyDrive/Classroom/Projects/Musi...	rock

MUSIC GENRE CLASSIFICATION¶

GROUP-1¶